Mesh Llm

Mesh LLM is software for distributed model inference. It serves large language models across several machines through a single OpenAI compatible endpoint, so a model too big for one GPU can be split into layer slices that run on right sized nodes across a homelab, a business network or the internet. Tools talk to a local endpoint, with the base URL set to localhost on port 9337, and the mesh handles placement. It has two modes. In router mode, the model field of the request selects which node hosts it, so many models can be served from one mesh. In split mode, a planner assigns ranges of layers from one large model to different machines, with activations passed between them over QUIC while the weights stay local. The site shows these working with tools such as goose, VS Code and OpenCode, and with any OpenAI client. A plugin architecture lets plugins declare what they provide in a manifest, with capabilities exposed over MCP, HTTP, inference and mesh events. A searchable catalog of Hugging Face models shows which ones ship layer packages for multi machine serving, and a node can be embedded in an application on supported platforms. The site offers a public mesh to join and a command line setup flow that begins with an installer script. It is aimed at people who run models on their own hardware and want to pool GPUs. The site publishes no pricing page or plan prices.